AI Potluck
Infrastructure / Deployment

OpenSandbox

Alibaba Cloud

Apache-2.0 sandbox runtime from Alibaba aimed squarely at AI agents, defines a 'Sandbox Protocol' with standard lifecycle/execution APIs and ships multi-language SDKs (Python, Java, JS, .NET) plus Docker and Kubernetes runtimes. Picked by teams wanting an open protocol-driven alternative to E2B at K8s scale; ~10.8k stars and 910 commits in the last 90d signal serious investment.

Alibaba OpenSandbox, general-purpose sandbox platform for AI agents (coding agents, GUI agents, agent eval, RL training); open sourced Mar 2026; components/execd 1.0.18 (May 25 2026) confirmed live on GitHub. Go execd daemon; Docker/Kubernetes runtimes; multi-language SDKs (Python/TS/Go/Java); MCP server. Self-host only (no managed tier observed).

Openness

5 high confidence
5.0
license
Apache-2.0(OSI)
source
public(Go execd + multi-language SDKs)
runtimes
Docker/Kubernetes
core-gated
ungated

Apache-2.0, fully self-hostable, no managed/enterprise gating observed => open_source (not open_core). Alibaba says it open sourced the sandbox infra used internally.

Adoption

2 low confidence
2.0

Only ~2-3 months old; no PyPI/download or named-deployment figures surfaced. ~11.3k GitHub stars (3,845 within 72h of launch) is the only signal, stars_fallback, which caps level at 3. Placed at 2: rapid early stars but no usage volume yet. Revisit as usage data emerges.

Capability

3 medium confidence
3.0

Broad multi-language SDK surface and MCP integration with K8s scale, but container-level isolation (no VM/microVM tier observed) keeps it mid-tier on the isolation-strength dimension vs Firecracker/gVisor/Kata. Strong breadth, moderate isolation.

Unchanged since 2026-07-30 (last edited, not re-checked)